Skip to content

ZenUml/conf-app / .claude/skills/customer-timeline

customer-timeline

Verified

Build the chronological timeline of every recorded fact for ONE Confluence tenant (Marketplace license and transactions, Mixpanel events, prod D1 tables, plus rows you pass in), with a source-coverage table and tracking-gap flags. Triggers on "完整时间线", "客户时间线", "customer timeline", "what did this customer do", "trial review timeline", "查这个租户所有数据源", "what happened on this site". Not for deciding whether to contact the customer (customer-followup), fleet-wide analysis (conf-app), or tracker writes (customer-tracker).

Customer timeline

One tenant in, one merged timeline out. The script does the mechanical sweep; you judge what the rows and gaps mean.

Input and output

Input: one identifier, a bare subdomain, a hostname, a cloudId, or a Marketplace entitlement id (E-xxx-xxx-xxx-xxx, or a license-details URL containing one). Optional --from / --to (UTC date or datetime; default: first Marketplace evidence to now), --tz <IANA zone> for the customer's local-time column (Australia/Melbourne is always added), and --extra rows.json (repeatable).

Output: rows {ts_utc, source, event, certainty (fact | derived), props} as JSON and as a markdown table, a coverage table (rows found (n) / queried, empty / not queried (reason) / unavailable (error) / missing), and gap flags. Files go to --out (default: a temp dir outside the repo). Customer data never goes into the repo; the script refuses --out inside a checkout unless it is under private/. Do not paste rows into PRs or issues.

Run

python3 .claude/skills/customer-timeline/scripts/timeline.py <identifier> [--from 2026-11-01] [--to 2026-11-17] [--tz America/New_York] [--extra tickets.json] [--out DIR]
python3 .claude/skills/customer-timeline/scripts/timeline.py <identifier> --plan     # identity and window only, no D1 or Mixpanel query
python3 .claude/skills/customer-timeline/scripts/test_timeline.py                    # unit tests, no network

Reads from the primary checkout when run in a worktree: marketplace.db, .env.mixpanel, wrangler.toml and node_modules are gitignored. Set CONF_APP_ROOT if it lives elsewhere. A missing file or credential becomes a named coverage gap, never a crash. --skip marketplace,mixpanel,d1 leaves a source out; --allow-scan lets it scan large unindexed D1 tables.

What the script reaches

SourceHowTraps it already handles
Marketplacemarketplace.db snapshot (mp_report.py sync refreshes it); license dates and transactions, no contactsOf all checkouts' snapshots the most recently synced is used; its file and sync time are in coverage. An identifier that is not found names that file, its age and the sync command: a stale snapshot is a gap, not "nothing matched". A site with no license (internal sites) falls back to D1 for its cloudId
Mixpanelmixpanel/scripts/mp_query.py --file, client_domain = bare subdomain, 14-day chunks, property subset only. A second pass (mixpanel-by-account) adds events with no client_domain sent by the accounts seen in pass one, as derived rows; skipped if there are none or more than 20Starts 2026-04-18. An account can belong to several sites, so treat derived rows as leads. Read the mixpanel skill before reading identity or sampling from the rows
D1 conf-zenuml-prodwrangler d1 execute ... --remote as in new-customers, one statement per table, tenant column found with pragma_table_info at run timeTable gone = missing. No usable index = not queried. Row cap 2000, TRUNCATED is reported. Bodies, secrets, ciphertext and emails are never selected. Staging sites (lite-stg, full-stg) have no prod D1 rows

What needs agent tools, then --extra

Support desk tickets (support-queue), the support mailbox, the tracker workbook (customer-tracker, read only), session replay (Mixpanel MCP Get-User-Replays-Data), and Stripe or Slack. Gather those rows yourself, write them as [{"ts_utc": ..., "source": "desk", "event": ..., "<anything else>": ...}], and pass --extra. They merge into the timeline and show in coverage. Say in your answer which of these you did not check.

Judgement steps

  1. Read coverage first. Quote any unavailable, not queried, missing or TRUNCATED source before drawing a conclusion. Empty is evidence only where the row says queried, empty.
  2. Treat gap flags as leads. sampled_events means absence proves nothing. start_without_terminal and silent_stretch point at an editor that lost its terminal hook or a user who got stuck; open the replay if authoring_without_replay does not list it. d1_vs_mixpanel_content is derived: each side is a lower bound.
  3. Mixpanel macro_viewed is a view of a macro, D1 page_updated is an editor action: do not add them. Keep fact and derived rows apart in the write-up.
  4. Never state a cause from a timeline alone. A cause needs a reproduction (global rule).

Fleet comparison and baselines: conf-app. Whether to contact the customer and what to say: customer-followup. Tracker updates: customer-tracker.

finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 197kUpdated Claude CodeCodexVerifiedInstall